Helping business gain AI insights on their data through voice search
tldr – what's the project about?
"Voice to Insight" is an machine learning NLP-based search feature where people could gain business insights from questions such as "What were the sales for Q3 2021?”…which would generate a chart.
role + timeline:
My role was devising the vision and interaction patterns for supporting a mobile-based query approach.
I led the design for the feature, and was shipped in 6 months alongside 1 product manager, 3~ developers, and my design manager.
success metrics:
An enhanced mobile experience
This contributed to the continuous delivery of desktop parity to mobile app features, and smart enhancements.
Brining a more mobile-friendly pattern:
On the desktop, users can type and directly select the appropriate model from suggestions. However, typing a complex question on mobile could be a bit cumbersome for a long question.
For our mobile users, the main use was to allow users to gain quick business insights on their data. Desktop users were generally more in-depth to create charts and datasets.
So while we already had the existing typing-based approach for our mobile app...the top customer request was to bring a voice flow.
Key Insight:
We need to reduce the friction of typing a long question on mobile...and instead replace with voice search to allow users to generate questions quicker
Activate a Search:
Clicking on the microphone button instantly activates a voice search. Users could then speak their results that would automatically generate a relevant chart.
Testing Different Interaction Methods:
Through the design process, different ways of activation were explored. But ultimately, integration of a simple microphone button allowed for a seamless approach to activate a search in real-time.
Clear Animation Interaction States
There are 3 stages to a voice search: beginning, listening to the search, and processing the query to produce a result. This was communicated through differences in speed in the animation.
Building a Fallback:
Sometimes voice input can fail. Users were able to recover by going back to a keyboard-based input if they needed to change a part of the query.
You can learn more about this in the above video, along with SAP's blog post.